Professional Documents
Culture Documents
03 Reliability of Generation Systems
03 Reliability of Generation Systems
as, for example: the required load level or the reliability of the individual
generating units. Clearly, the higher the reliability level of the individual
units, the higher will be the reliability level of the whole generation system.
Generally speaking, the reliability level of any system is determined by
the risk that under certain conditions the system will not operate
satisfactorily due to failures in the system elements. In the generation
system these indices can be determined by a relatively simple approach,
namely by comparison of the required load to the generation capacity in the
probabilistic manner.
As in an entire system, its generation part reliability is described with the
aid of two types of such risk indices. The first type includes the risk
relating to the static condition of the system, i.e., the system behavior at
steady state. The second type includes the risk for the dynamic conditions,
i.e., the system behavior under transient phenomena. Thus, the static
indices do not relate to the transient phenomenon starting at the failure
occurrence and ending at the new steady state condition. If, for example,
the required reserve is studied, one should differentiate between operational
reserve (which includes the spinning or hot reserve and a quick start of gas
turbines) and the installed reserve capacity. The operational reserve relates
to the dynamic condition of the system, i.e., to the system operation, while
the installed reserve relates to the system design which determines the
number and size of the units and the time of their operation in response to
increasing electricity demand.
The static and dynamic system conditions can be illustrated by the
following example. Assume that one of the large generating units is out of
operation due to a failure, when the load level is, say, at peak. At that
moment, the system enters a transient condition, where the controls of
frequency react immediately. If the frequency rate drops or the frequency
itself reaches certain predetermined values, a load shedding procedure (i.e.,
disconnection of certain consumers) takes place. Such a procedure may
take a few seconds. At that point, the dispatcher intervenes by, for example,
“throwing in” of gas turbines, which takes minutes to start operating. Later,
if required, hot units can be committed in the system. This may take hours.
Finally, the power system will reach a new steady state, where there may
still be unsupplied consumers due to shortage in generating capacity.
Two distinct states characterize the above procedure, namely:
1. The dynamic state, which starts at the moment of failure and
continues throughout the procedure of adding generating units,
2. The static state, which is the new steady state.
The reliability indices for the generation system at steady state are
discussed first. The installed reserve, which is required to provide the
Reliability of Generation Systems 173
This index is not probabilistic. It has been used until new models based
on probabilistic reliability indices were developed. As for the reserve, it
was determined by accumulated praxis. It should be noted that the RM
index does not take into account the type of the generating units (thermal,
gas turbines, etc.), their sizes and their reliability levels. Thus, two systems
with markedly different reliability levels may have the same RM value. As
an example, consider two systems with the same annual peak load value
and the same RM. The first system is composed of large generating units
and the second of smaller units, say 50% of the large units. It is obvious
that the second system is characterized by higher reliability level. The
reason for that is that the probability for one large unit outage is higher than
the probability for a simultaneous outage of two small units.
Another disadvantage of the deterministic indices is that they ignore the
stochastic nature of factors influencing the reliability level, like, for
example, the available generation capacity or the predicted load level. As a
result, the deterministic indices overestimate the system reliability in
contrast to a more sound probabilistic approach.
Accurate reliability evaluation should take into account the unit
commitment and given operational constraints. The combination of a
thorough mathematical analysis with computer-based computations gives
not only more realistic indices values by simulating the unit commitment
through randomly generated availabilities of the units but also allows for
174 New Computational Methods in Power System Reliability
The calculation of reliability indices for the steady state, required for the
planning of the generation system is based nowadays on two analytical
methods. These methods enable to determine the required reserve for the
generation system, taking into account the predicted load behavior and a
predetermined reliability level. The first method is the loss of load proba-
bility, which means the probability of the load being larger than the
available generation. The second method enables, besides the determi-
nation of the probability, to calculate the frequency and duration where the
load is larger than the available generation. In this method, the frequency
and duration, besides the probability, are studied. It is called the Frequency
and Duration Method. The main difference between these two methods lies
in the load models and the reliability parameters of generating units.
Both methods mentioned above are widely used, though the first one is
more “popular”. It should be mentioned that there is a third method for
calculation of reliability indices, based on Monte Carlo simulation. All
these methods are presented and discussed in this chapter.
G
G1 ~
G2 ~ L
:
Gn ~
Fig. 3.1. The simple model of the generation system and the load (G1,..., Gn are
nominal generating capacities of n generating units, L is the load at a certain time)
The LLF produces a binary value for any combination of G and L. The
expected value of the random binary LLF is equal to the probability that
LLF=1 or, in other words, to the probability of the generation system
inability to supply the load. This index is usually referred to as LOLP (Loss
of Load Probability):
LOLP(G, L) = E{LLF(G, L)} = Pr{LLF(G, L) = 1} = Pr{G < L} . (3.3)
The total available generation capacity G usually takes on discrete values
and, therefore can be represented by a pmf gj, pj=Pr{G=gj} for 1≤j≤n. The
discretized load curve is also represented by its pmf li, qi=Pr{L=li} for
1≤i≤k. Having the pmf of the two random variables one can obtain the pmf
of their function LLF(G,L). For this purpose the pmf of G and L should be
represented by u-functions
n gj
U G ( z) = ∑ p jz (3.4)
j =1
and
k
U L ( z) = ∑ qi z l i . (3.5)
i =1
176 New Computational Methods in Power System Reliability
Applying the composition operator ⊗ over UG(z) and UL(z) one obtains
LLF
The Loss of Load Expectation (LOLE) represents the expected time (in
hours or days) where the load is larger than the available generation.
Assume that the operation time T is divided into S equal intervals δt=T/S
(time units). In each interval s the load takes on a random value of L (it is
assumed that during the interval the load does not vary). The random time
when the constant load L is not supplied in the interval s is equal to
τs=δt⋅LLF(G,L). This variable can take on either value of δt or 0. The
random time when the variable load is not supplied during the entire
operation period T can be obtained as
S S S T T S
∑τ s = ∑ δt ⋅ LLF (G, L) = ∑ ⋅ LLF (G, L) = ∑ LLF (G, L). (3.8)
s =1 s =1 s =1 S S s =1
n gj k
UUL ( z ) = U G ( z ) ⊗ U L ( z ) = ( ∑ p j z ) ⊗ ( ∑ qi z li )
UL j =1 UL i =1
(3.11)
n k max{li − g j ,0}
= ∑ ∑ p j qi z .
j =1i =1
Example 3.1
Consider two separate generating systems (power plants) with a nominal
capacity of 100 MW. In the first system, some types of failure require its
capacity G1 to be reduced to 60 MW and other types lead to a complete
outage. In the second generator, some types of failure require its capacity
G2 to be reduced to 80 MW, others lead to a capacity reduction to 40 MW,
and the rest lead to a complete outage. The generators are repairable and
each of their states has a steady-state probability.
Both generators should supply a variable two-level load L. The high
level (day) load is 50 MW and has the probability 0.6; the low level (night)
load is 30 MW and has the probability 0.4.
The capacity and the load can be presented as a fraction of the nominal
generator capacity. There are three possible relative capacity levels that
characterize the performance of the first generating system:
g10 = 0.0, g11 = 60/100 = 0.6, g12 = 100/100 = 1.0, and four relative
capacity levels that characterize the performance of the second generating
system:
g20 = 0.0, g21 = 40/100 = 0.4, g22 = 80/100 = 0.8, g23 = 100/100 = 1.0
Assume that the corresponding steady-state probabilities are
p10 = 0.1, p11 = 0.6, p12 = 0.3
for the first generating system and
p20 = 0.05, p21 = 0.35, p22 = 0.3, p23 = 0.3
178 New Computational Methods in Power System Reliability
for the second generating system and that the load distribution is
l1 = 50/100 = 0.5, l2 = 30/100 = 0.3, q1 = 0.6, q2 = 0.4.
The probabilistic distributions of the generation capacities and the load
are presented in Table 3.1.
0 0.6
= (0.1z + 0.6 z + 0.3z1) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
LLF
= 0.06 z1(0<0.5) + 0.36 z1(0.6< 0.5) + 0.18 z1(1< 0.5) + 0.04 z1(0< 0.3)
+ 0.24 z1(0.6< 0.3) + 0.12 z1(1< 0.3) = 0.06 z1 + 0.36 z 0 + 0.18 z 0 + 0.04 z1
+ 0.24 z 0 + 0.12 z 0 = 0.9 z 0 + 0.1z1
Reliability of Generation Systems 179
U LLF 2 ( z ) = U G 2 ( z ) ⊗ U L ( z )
LLF
0 0.4
= (0.05 z + 0.35 z + 0.3z 0.8 + 0.3z1) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
LLF
UUL 2 ( z ) = U G 2 ( z ) ⊗ U L ( z )
UL
= (0.05 z 0 + 0.35 z 0.4 + 0.3 z 0.8 + 0.3 z1 ) ⊗ (0.6 z 0.5 + 0.4 z 0.3 )
UL
One can obtain the u-function representing the pmf of the entire system
capacity UG(z) applying the following operators over u-functions UGj(z)
representing the pmf of the units’ capacities:
U G ( z ) = U G1 ( z ) ⊗U G 2 ( z ) ⊗ ... ⊗U G n ( z ). (3.15)
+ + +
Example 3.2
Consider a generation system consisting of three generating units with
variable random capacity (the capacity variations are caused by failures and
maintenance/inspection actions). The system should supply a variable two-
level load L. The high level (day) load is 260 MW and has the probability
0.6; the low level (night) load is 180 MW and has the probability 0.4. The
distributions of unit capacities and of the load are presented in Table 3.2.
The u-functions representing the capacity distribution of the generating
units (the pmf of random variables G1, G2and G3) take the form
UG1(z) = 0.02z0+0.98z50, UG2(z) = 0.02z0+0.02z50+0.96z100,
UG3(z) = 0.03z0+0.01z50+0.02z100+0.94z150.
Reliability of Generation Systems 181
Table 3.2. Distributions of the unit generation capacities and the load
State 1 State 2 State 3 State 4
First generating g (MW) 50 0 - -
unit p 0.98 0.02 - -
Second g (MW) 100 50 0 -
generating unit p 0.96 0.02 0.02 -
Third generating g (MW) 150 100 50 0
unit p 0.94 0.02 0.01 0.03
Load l (MW) 240 110 - -
q 0.6 0.4 - -
+0.94z150)=10-4(0.12z0+6.04z50+13.72z100+293.88z150+289.84z200
+552.88z250+8843.52z300).
The expected steady-state capacity of the generation system can be
obtained directly from the u-function representing the pmf of the system
capacity:
E(G)=U'G(1)=10-4(0.12×0+6.04×50+13.72×100+293.88×150
+289.84×200+552.88×250+8843.52×300)=289.5
The u-function corresponding to the pmf of the LLF are obtained using
the composition operator:
U LLF 1 ( z ) = U G ( z ) ⊗ U L ( z )
LLF
= 0.6 ⋅ 10-4 (0.12 z1(0< 240) + 6.04 z1(50< 240) + 13.72 z1(100< 240) + 293.88 z1(150< 240)
+ 289.84 z1( 200< 240) + 552.88 z1( 250< 240) + 8843.52 z1(300< 240) )
+ 0.4 ⋅ 10-4 (0.12 z1(0<110) + 6.04 z1(50<110) + 13.72 z1(100<110) + 293.88 z1(150<110)
+ 289.84 z1( 200<110) + 552.88 z1( 250<110) + 8843.52 z1(300<110) )
182 New Computational Methods in Power System Reliability
-4 240 190
= 0.6 ⋅ 10 (0.12 z + 6.04 z + 13.72 z140 + 293.88 z 90
+ 289.84 z 40 + 552.88 z 0 + 8843.52 z 0 ) + 0.4 ⋅ 10-4 (0.12 z110 + 6.04 z 60
+ 13.72 z10 + 293.88 z 0 + 289.84 z 0 + 552.88 z 0 + 8843.52 z 0 ).
and
T
EUE(x) = ∑ EUE ( x(t )) . (3.17)
t =1
Here EUE(x(t)) represents the energy not supplied at the instant t for
scenario x and
⎧1 if EUE ( x(t )) > 0
LOLP(x(t)) = ⎨ (3.18)
⎩ 0 otherwise
The expected value of loss of load probability is calculated as:
LOLP = ∑ LOLP( x) ∗ P( x) , (3.19)
x∈ X
and
1 M
EUE = ∑ EUE ( x m ) , (3.22)
M m =1
and
1 M
V(EUE) = ∑ ( EUE ( x m ) − EUE ) 2 (3.26)
M m =1
and
β EUE = V ( EUE ) / M / EUE (3.30)
⎧⎪ V ( LOLP ) V ( EUE ) ⎫⎪
M = max ⎨ , ⎬ (3.31)
⎪⎩ ( β ∗ LOLP ) ( β ∗ EUE ) 2 ⎪⎭
2
The sets S2 and S3 consist of K' states. Each state is defined by a capacity
level from 0 to nominal (not including nominal as it belongs to S1). Thus, a
generating unit can be put in one of the K=2*K'+1 states. For the
probability simulation of generation system operation it is assumed that
transition from one availability state to another may only depend on the
assumed current state and belong to the corresponding time subperiod PTτ.
In such a way a generating unit state transitions can be modeled by Markov
process.
Let us denote the probability of transition of unit n to state j from state i
in time subperiod PTτ by pnτij. For each unit n there are two transition
matrices Pnτ corresponding to the subperiod PTτ:
⎛ pnτ 11 pnτ 12 … pnτ 1K ⎞
⎜ ⎟
⎜p pnτ 22 … pnτ 2 K ⎟
Pnτ = ⎜ nτ 21 (3.32)
… … … … ⎟
⎜ ⎟
⎜p pnτK 2 … pnτKK ⎟⎠
⎝ nτK1
These matrices are known as the stochastic transitional probability
matrices. It should be noted that the sum of all the probabilities in each row
of the matrix must equal 1 since each row i represents all the possible
transitions to the other states in a particular instant of time given that it is in
the state i.
For a class of Markov chain possessing the ergodic property, the state
probabilities will converge to a vector of steady-state probabilities
pnτ = (pnτ1 pnτ2 … pnτK) (3.33)
such that
pnτ * Pnτ = pnτ. (3.34)
This matrix equation together with
K
∑ pnτk = 1 (3.35)
k =1
subperiod PT1). The initial state k0 of the generating unit is equal to k(x) To
sample a generating unit state in a certain instant of time the same sampling
procedure as in the initial instant is performed first to determine the value
x. Next the index k(x) is found to satisfy the inequality
k ( x ) −1 k ( x)
∑ pnτk s k < x ≤ ∑ pnτk s k (3.37)
k =1 k =1
in the corresponding subperiod PTτ for the current state ks. The next unit
state kn is equal to k(x). Execution of the sampling procedure for each
instant of time during the study period results in the unit availability
scenario.
The set Tn of instants at which kn≠ks and the available unit capacity has
changed divides the study period into time intervals. The state of the unit,
regarding the unit commitment during each interval, does not change. Set
Tn contains the set Tn of instants at which kn∈ S2 (the instants of failures)
and the available unit capacity decreased. The set Tn divides the study
period into time intervals so that the unit state can vary during these
intervals but these variations are known beforehand and, therefore, it is
possible to take them into account in unit commitment procedure.
Joining up all generating unit availability scenarios leads to the entire
generation system availability scenario. The set
N
T= ∪ Tn (3.38)
n =1
divides the study period into time intervals so that the variations of the
system state during each time interval are known from the beginning (there
are no failures during this interval). Consequently, it is possible to fulfill
the unit commitment for such an interval entirely.
Now we have a sequence of the generation system operation scenarios.
Each scenario consists of integration of generating unit operation scenarios
and subdivision of the study period into the time intervals without failures.
The purpose of unit commitment is to define, for each unit, the on line
intervals during the unit commitment period to meet the load demand and
to satisfy the generation system operation policy. The unit commitment
period is subdivided into time intervals of the same length. This length is
190 New Computational Methods in Power System Reliability
equal to the minimum time in order to start up an emergency unit (as a rule
a jet turbine). Load is represented by peak loads at these time intervals.
The spinning reserve requirements are given in percent of load demand
or in MW and may vary during the study period. It should be noted that
sometimes it is impossible to satisfy the spinning-reserve requirements
because of the generating units capacity shortages. These constraints can be
written in the form
∑ OC jt ≥ LOADt + SRt (3.39)
j∈U t
If τ0>0 then unit j is on line in interval [1,t0) and off line otherwise.
Based on the commitment priority rule, unit j should be committed at a
free-decision instant t∈ [t0,T] if there is unsupplied system demand at this
instant ( SLjt > 0 ). As a result, each free-decision instant t must be either of
the following:
• An on line instant (SLjt > 0 ),
• An unclassified instant ( SLjt ≤ 0 ) , for which commitment status has not
yet been determined
Let Ik, k = 1,...,K represent the closed interval k of consecutive
unclassified instants contained in the free decision interval [t0,T], and let Tk
represent the duration of interval Ik. These unclassified intervals are
numbered according to their chronological orders. Two adjacent
unclassified intervals are separated by an on line interval. Let Ik = [tk1,tk2] ,
then
Tk = tk2 - tk1 + 1. (3.41)
Let τk represent the duration of interval when unit j was on-line (τk >0) or
off-line (τk<0) just before the instant tk1. Based on τk, the tentative
commitment strategy associated with unclassified interval Ik can be
estimated as follows:
j
• If 0 < τk < TMUP - Tk then unit will be set on-line during the whole
interval Ik;
j j
• If 0 < τk < TMUP and τk > TMUP - Tk then determine
j j
tk1 = tk1 + (TMUP − τ k ) , Tk = Tk − (TMUP −τ k )
j
− If Tk < TMDN then the unit will be set on line during the whole
interval Ik;
j
− If Tk ≥ TMDN then the unit will be set on line during the interval
[tk1, tk1 ) and off-line during the interval [ tk1 ,tk2]
192 New Computational Methods in Power System Reliability
j j
• If τk ≥ TMUP and Tk < TMDN then the unit will be set on-line during the
whole interval Ik;
j j
• If τk ≥ TMUP and Tk ≥ TMDN then the unit will be set off-line during the
whole interval Ik;
Example 3.3
Using a very simple example we show how to model power system
operation for obtaining a statistically reliable assessment of the loss-of-load
probability and the unserved energy. The random nature of the forced
outages is described by the Markov chains combined with the Monte-Carlo
simulations of the unit availability.
The generating system is formed of only five units detailed in Table 3.3
The units are sorted in the ascending order of their power generating
cost, that is, the first unit produces the cheapest energy. This sorting
defines the merit order for unit commitment.
We also suppose that each unit can be in either one of three different
states: full unavailability (i=1), 50% availability (minimum capacity) (i=2)
and full availability (i=3). The (3x3) probability transition matrices are
presented in Table 3.4, where Pni stands for the probability of the unit
transition from the n-th to i-th state every 15 minutes. The diagonal entries
of the matrices dominate the off-diagonal ones due to the unit ability of
preserving its current state. The transition diagram, corresponding to the
first matrix is shown on Fig. 3.2.
Unit 4 Unit 5
Pni i=1 i=2 i=3 i=1 i=2 i=3
n=1 98% 1% 1% 93% 3% 4%
n=2 2% 97% 1% 4% 91% 5%
n=3 3% 2% 95% 5% 3% 92%
0.02
0.95 0.94
0.97
0.03 0.02
500 Mw 250 Mw 0 Mw
0.01
0.04
0.02
For reader convenience the partial and full unavailability of the units are
boldfaced.
At first glance, the total availability of the units permits to provide
LD+SR at each moment. However, in actual practice, the more expensive
units (4 and 5 in our example) are shut down, whenever possible, to
minimize the energy generation cost. This policy may lead to the SR
shortage or, more severely, to the unsupplied energy, when either of the
basic units 1-3 is not fully available due to an unexpected failure. Indeed,
the unit 4 cannot be started immediately (because of its relatively long start
up time) while the quickly started unit 5 has too small capacity for
compensating even a partial unavailability of the basic units.
Tables 3.6 and 3.7 display the corresponding dispatch and spinning
reserve contributions of the units.
At the first moment all the units are fully available. The generation of the
required LD = 900 Mw is provided by nominal loading of the two cheapest
units and partial loading (200 Mw) of the third unit. The rest of its available
capacity (300-200) = 100 Mw covers the required SR = 50Mw. The
expensive units 4 and 5 are not started up at all. At the moment (Time =
15) unit 2 drops down to only 50% of its nominal capacity 100 Mw), so
that the total available on-line capacity at this moment is (500+100+300) =
900 Mw. This is sufficient only to satisfy the load demand (900 Mw) with
no spinning reserve contribution.
196 New Computational Methods in Power System Reliability
Time Units
(Minutes) 1 2 3 4 5 Total
0-15 0 0 100 50
15-30 0 0 10 10 20
30-45 0 0 50 50 100
45-60 0 0 50 50 100
60-75 0 0 50 50 100
75-90 0 0 150 50 200
90-105 0 0 100 100
105-120 0 0 100 100
120-135 0 0 100 100
135-150 0 0 0 0 0
150-165 0 0 0 0 0
165-180 0 0 0 0 0
180-195 0 0 150 50 200
195-210 0 0 150 50 200
210-225 0 0 100 50
225-240 0 0 100 50
start up unit 5 with practically zero start-up time at minimum loading level
(10Mw) and to reduce loading of unit 3 to 290 Mw. This operation
contributes only 20 Mw to the spinning reserve.
The shortage (50-20) = 30Mw of spinning reserve forces the dispatcher
to start up unit 4 which becomes available after 15 minutes, which are
needed for the start-up, at the moment (Time = 30). Unit 4 is loaded at the
minimal level of 50 Mw and unit 3 is loaded at the level of 250 Mw.
Spinning reserve contribution will be 100 Mw.
It continues to work up to the moment (Time = 75), when unit 2 restores
its full availability. For saving purposes, unit 4 should be shut down.
Nevertheless, because of the minimum up time requirement it will continue
to be on line for another 15 minutes up to the moment (Time = 90). During
this period unit 2 is loaded at the nominal capacity and unit 3 at the level of
150 Mw. Spinning reserve contribution will be 200 Mw.
At the moment (Time = 90) unit 4 is shut down and unit 3 is loaded at
the level 200 Mw. Spinning reserve contribution will be 100 Mw.
At the moment (Time = 135), unit 3 possesses only 50% of its
availability. Again, as before, unit 4 can not be started up immediately.
Instead, the dispatcher performs the quick start of unit 5. This gives totally
(500+200+150+20) = 870Mw, with (900-870) = 30Mw unsupplied.
At the moment (Time = 150) unit 4 is started but unit 2 drops to the level
of 100 Mw. With no regard to ramp rate requirement let us suppose that the
load of unit 4 could be immediately increased to the nominal level.
Therefore, the total available capacity is (500+100+150+100+20) =
870Mw, with (900-870) = 30Mw unsupplied up to the moment Time =
180, when units 2 and 3 will be available.
Due to the minimum down requirements unit 4 is not shut down until the
moment Time = 210. Thus, the load of unit 2 is increased to the nominal
capacity, the load of unit 4 is decreased to the 50Mw, unit 5 is shut down
and the load of unit 3 is not changed. Spinning reserve contribution is then
200 Mw.
At the moment Time = 210 unit 4 is shut down and unit 3 is loaded at the
level 200 Mw. Spinning reserve contribution is equal to 100 Mw.
In this example we have 45 minutes of loss of load = 0.75LOLH
(Time135-180) and the unsupplied energy (30Mw*0.75) = 22.5MWh.
198 New Computational Methods in Power System Reliability
GU1,1 GU2,1
... ...
GUM-1,1 GUM,1
GUM-1,N GUM,N
contingency states, AGi is less than a load of the MSGS. We assume that
{ 1,2,..., m }∈NS, and { m + 1,..., K g }∈ CS.
g
g g
200 New Computational Methods in Power System Reliability
MSRP1 MSTS1
Reserve
…
Reserve
MSRPM MSTSM
AG mi ≤ CRm so that
r
For the infirm contract, CRm is not guaranteed. It depends on the load
level of MSRP m for interval n Lnm . ARmi is defined as follows:
r
states.
The number of different states for the entire CGS is:
M M
K = K g ∏ K mr ∏ K ht (3.44)
m =1 h =1
If all the elements in the system are independent, the probability of state
i for the entire CGS is determined by the state probabilities of MSGS,
MSRPs and MSTSs associated with state i:
Reliability of Generation Systems 201
M M
pi = pi g
∏ p mi ∏ p mi
r t
(3.45)
m =1 m =1
When the CGS stays in state i, the performance level wi of the MSGS
after considering the assistance from the MSRPs can be determined based
on AGi , ARmir and ATmit .
g
∑∑ Pr{L }
N K
LOLP = n
> wi ⋅ Tn T (3.46)
n =1 i =1
∑∑ Pr{L }
N K
EENS = n
> wi ⋅ ( Ln − wi ) ⋅ Tn (3.47)
n =1 i =1
Problem formulation
The reserve contract between the MSGS and MSRP m is determined by the
contract type, CRm , per unit costs of reserve capacity and of the utilized
reserve provided by MSRP m ( ε Cm and ε Um respectively), and also by the
reliability of MSRP m and MSTS m. The objective of reserve management
for a MSGS is to determine the optimal reserve structure, which includes
the reserve contracts with MSRPs and the utilization order of MSRPs,
under the minimum total reserve cost while satisfying the system reliability
requirement. The total reserve cost includes the reserve capacity cost CC
and the expected reserve utilization cost EC. The reserve optimization
problem can be summarized as:
202 New Computational Methods in Power System Reliability
M
Min ∑ (CCm + ECm ) (3.48)
m =1
n
URmi {
≤ min ARmir , ATmit ; } (3.52)
Load curtailment LLni is utilized when the total generating and reserve
capacity cannot meet the load of the MSGS.
[ ]
GN
U MSGS ( z ) = ⊗
+
U G1 ( z ),..., U G N ( z ) = ∏U j ( z) =
j = G1
(3.56)
[ ]= ∑ p
Kg
∏ (1 − A j )⋅ z + A j ⋅ z
GN AGig
0 w j ,1
ig ⋅z
j = G1 i g =1
where A j and w j ,1 are the availability and the nominal generating capacity
of generating unit j respectively.
Notice that the first mG terms in the U MSGS (z ) with the associated states
belonging to NS can be removed from the u-function without changing the
results of reliability indices. In order to simplify the algorithm one can
remove the first mG terms from the u-function and define it as
Kg
AGi g
U MSGS ( z ) = ∑ p ig ⋅ z . (3.57)
i g = m g +1
Although all the generators of a MSRP are also connected in parallel, the
u-functions representing the distribution of the random reserve depend on
the type of the reserve contracts. For MSRP m with Rm N generating units
which are connected in parallel, the u-functions of MSRP m for the firm
and infirm contracts can be represented using the operators ⊗ and ⊗ ,
firm infirm
[( ]
⎧⎪ Rm N ⎫⎪ (3.58)
w
)
= ⊗ ⎨ ∏ 1 − A j ⋅ z 0 + A j ⋅ z j ,1 , CRm ⎬
firm ⎪ j = Rm
⎩ 1
⎪⎭
K mr K mr
min( AG mir , CR m ) AR
= ∑ pmir ⋅ z = ∑ pmi ⋅ z mir .
r
i r =1 i r =1
[
U MSRP (Z ) = ⊗ U Rm1 ( z ),...,U RmN ( z ), CRm , Lnm =
infirm
]
⎡ RmN ⎤
⎡
( ⎤
)
= ⊗ ⎢ ⊗ U Rm1 ( z ),...,U RmN ( z ) , CRm , Lnm ⎥ = ⊗ ⎢ ∏U j ( z ), CRm , Lnm ⎥
infirm⎣infirm ⎦ infirm⎣⎢ j = Rm1 ⎦⎥
(3.59)
⎧⎪ RmN
[( w
) ⎫⎪
= ⊗ ⎨ ∏ 1 − A j ⋅ z 0 + A j ⋅ z j ,1 , CRm , Lnm ⎬
infirm⎪ j = Rm ⎪⎭
]
⎩ 1
r r
N Km min(max(AGmir − Lnm ,0), CRm ) N Km ARmir
= ∑ ∑ pmir ⋅ z ⋅ Tn / T = ∑ ∑ pmir ⋅ z ⋅Tn / T
n =1ir =1 n =1i r =1
[ ]
Tm N
U MSTS ( Z ) = ⊗ U Tm1 ( z ),..., U Tm N ( z ) =
+
∏U j ( z)
j = Tm1
(3.60)
Tm N
j = Tm1
[
∏ (1 − A j )⋅ z + A j ⋅ z 0 w j ,1
]= ∑ p
K mt
it =1
mit ⋅z
ATmit
WR
The u-function of the MSGS U MSGS ( z ) with the reserve agreements with
MSRPs and the capacity limits of MSTSs can be now obtained as:
WR
U MSGS { [
( z ) = ⊗ U MSGS ( z ),..., ⊗ U MSRS m ( z ), U MSTS m ( z ) ,..., ]
]}
+ min
(3.62)
[
⊗ U MSRS M ( z ), U MSTS M ( z )
min
WR
From U MSGS ( Z ) we can obtain the LOLP using the following equation
N K
{ }
LOLP = ∑∑Pr Ln > wi ⋅Tn T = ∑ ∑ pi ⋅ 1(Ln − wi )⋅Tn T
N
(3.63)
n=1i=1 n=1i∈CS
Reliability of Generation Systems 205
where CRm takes the values in the interval [CRmmin , CRmmax ] and k is the
decimal number represented by the binary substring.
Using this encoding algorithm, the solutions for obtaining the reserve
utilization order are within the feasible space. According to the string
represented in Fig. 3.5, MSRP 2 is used first, MSRP 1 is used second, and
so on, up to the point the load is met, or the whole available reserve is used
in a contingency state.
To let the GA look for the solution with minimal total cost without
violating the constraints LOLP ≤ LOLP spec and EENS ≤ EENS spec , the
solution quality (fitness) is defined as
M
F = q ⋅ max(0, LOLP − LOLP spec , EENS − EENS spec ) + ∑ (CCm + ECm ) (3.66)
m =1
M S R P1 M S R P 2 M S R P 3 M S R P1 M S R P 2 M S R P 3
u tiliz a tio n s e q u e n c e o f c o n tr a c tu a l r e s e r v e
reserve c a p a c ity
Example 3.4
Consider the restructured IEEE-RTS (IEEE Task Force 1979) consisting of
six MSGSs in which the optimal reserve structure should be determined for
the specific MSGS, which has reserve agreements with other five MSGSs.
The MSGS owns 6×50MW hydro units, 5×12MW oil units, 2×155MW
coal units and 2×400MW nuclear units. MSRP1 owns 2×20MW gas
turbine units and 2×76MW coal units. MSRP 2 owns 3×100MW oil units.
MSRP 3 owns 2×20MW gas turbine units and 2×76MW coal units. MSRP
4 owns 3×197MW oil units. MSRP 5 owns 2×155MW coal units and one
350MW coal unit. The MSRPs are connected to the MSGS through tie
lines.
The tie line reliability data (Allan et al. 1986) is used in this paper. We
assume that all the tie lines are identical, each with the capacity rating of
300MW and unavailability 0.00130873. MSTSs 1, 2 and 3 each have one
tie line. MSTSs 4 and 5 each have two tie lines.
The four-step model of the LDC obtained from hourly loads is used. The
peak load of the MSGS is 1253MW. For the MSGS, the load levels Ln and
the total time periods T n of a particular step are shown in Table 3.8. The
load level Ln of a particular step is determined by the average value of
hourly load in the time period of T n .
All the MSRPs are firm contract providers for the MSGS. The utilization
prices for MSRPs are 25$/MWh, 55$/MWh 75$/MWh, 60$/MWh and
20$/MWh, respectively. The reserve capacity prices for MSRPs are
2.5$/MWh, 5.5$/MWh 7.5$/MWh, 6$/MWh and 2$/MWh, respectively.
The time frame for the contracts is one year. Two examples with different
reliability constraints are investigated in this case.
Reliability of Generation Systems 207
In example A, the LOLP and EENS constraints for the MSGS are set as
0.01 and 500MWh/yr respectively. In example B, the reliability constraints
are tightened and the LOLP and EENS for the MSGS are 0.0005 and
50MWh/yr respectively.
Table 3.9 shows the total reserve cost and the reserve capacities between
the MSGS and MSRPs for the two examples.
Table 3.9. The total reserve cost and reserve structure for for the first case
Example A Example B
Total Reserve Cost (k$) 10197.0 10896.1
CR MSRP1 (MW) 53.728 70.355
CR MSRP 2 (MW) 1.7895 0.0185
CR MSRP 3 (MW) 0.2981 0.0186
CR MSRP 4 (MW) 6.4949 0.0524
CR MSRP 5 (MW) 374.82 419.02
The capacity reserve cost and expected cost of reserve utilization for
each MSRP are shown in Table 3.10. It can be seen from Tables 3.9 and
3.10 that the reserve cost has increased dramatically when the reliability
requirements were tightened.
Table 3.10. The capacity reserve cost and expected cost of utilized reserve for the
first case
Example A Example B
MSRP CC (k$) CC m (k$)
No. m EC m (K$) EC m ( K$)
u h ( z ) = A h z g h + (1 − A h ) z 0 = A h z g h + A h z 0 , (3.67)
where Ah is the availability of component h. The availability of a
component at steady state is calculated by two statistical parameters: the
Mean Time to Failure (MTTF) and the Mean Time to Repair (MTTR).
MTTFh µh
Ah = = (3.68)
MTTFh + MTTRh µ h + λh
where λh and µh are the forced outage rate and the repair rate of component
h respectively, λ = 1/MTTF, µ = l / MTTR.
The unavailability Ah which is also called FOR (Forced Outage Rate),
is calculated as:
λh
Ah = P(Gh = 0) = FOR = . (3.69)
µ h + λh
When component performance is defined as capacity or productivity, the
total capacity G{i, j} of a subsystem containing two independent
Reliability of Generation Systems 209
The u-function representing the pmf of G{i, j} can be, therefore, obtained
using the operator ⊗ over the u-functions representing pmf of Gi and Gj.
min
= ( A1z g1 + A1z 0 ) ⊗ ( A2 z g 2 + A2 z 0 )
min (3.71)
min{ g1 , g 2 } 0 0 0
= A1 A2 z + A1 A2 z + A1 A2 z + A1 A2 z
= A1 A2 z min{g1 , g 2 } + (1 − A1 A2 z 0 ).
where
U1,2 ( z ) = u1( z ) ⊗ u2 ( z ) and U 3,4 ( z ) = u3 ( z ) ⊗ u4 ( z ) .
min min
The u-function representing the pmf of G{i, j} can be, therefore, obtained
using the operator ⊗ over the u-functions representing pmf of Gi and Gj.
+
As in the case of the series connection, the u-functions for any parallel
system can be obtained recursively by consecutively determining the u-
functions of pairs of the components.
An example of a typical block diagram for a power unit is presented in
Fig. 3.6. In this figure the random capacities of different facilities,
subsystems and an entire power unit are determined in accordance with the
block diagram.
G11
G 31 G 32
G 12 G 22
G12
G 33 … G 1n
… G 23 G 24
G 34 G 25
G1k
⊗ operators.
+
3. Replace in the diagram the subsystem with single component having the
u-function obtained for the given subsystem.
4. If the reliability block diagram contains more then one element return to
step
The resulting u-function represents the capacity distribution of the entire
system.
Example 3.5
In order to illustrate the recursive approach (the reliability block diagram
method), consider the series-parallel power station coal feeding system
presented in Fig. 3.7A. The parameters of the system elements are
presented in Table 3.11.
First, one can find a series subsystem consisting of elements with the u-
functions u2(z), u3(z) and u4(z). By calculating the u-functions
U1(z)= u2 ( z ) ⊗ u3 ( z ) and U2(z)= U1( z ) ⊗ u4 ( z ) and replacing the three
min min
elements with a single element with the u-function U2(z) one obtains a
system with the structure presented in Fig. 3.7B. This system contains a
purely parallel subsystem consisting of elements with the u-functions U2(z)
and u5(z), which in their turn can be replaced by a single element with the
u-function U 3 ( z ) = U 2 ( z ) ⊗ u5 ( z ) (Fig. 3.7C). The structure obtained has
+
u7(z) u7(z)
A B
u7(z) u7(z)
C D
=0.504z3+0.496z0
U3(z)=U2(z) ⊗
+
u5(z)=(0.504z3+0.496z0) ⊗
+
(0.6z3+0.4z0)
= 0.3024z6+0.4992z3+0.1984z0
⊗ U3(z)=(0.9z5+0.1z0) ⊗ (0.3024z6+0.4992z3+0.1984z0)
U4(z)=u1(z) min min
= 0.27216z5+0.44928z3+0.27856z0;
5 3
⊗ u6(z)=(0.27216z +0.44928z
U5(z) = U4(z) min
+0.27856z0) min 6 0 5 3 0
⊗ (0.8z +0.2z )= 0.217728z +0.359424z +0.422848z ;
U(z)=U5(z) ⊗
+
u7(z)=(0.217728z5+0.359424z3+0.422848z0) ⊗
+
(0.8z3
+0.2z0)=0.1741824z8+0.2875392z6+0.0435456z5
+0.4101632z3+0.0845696z0.
Reliability of Generation Systems 213
Having the system u-function that represents its capacity distribution one
can easily obtain the system mean capacity E(G)=U'(1)=4.567. The system
LOLP for different constant load levels l can be obtained by summing the
probabilities corresponding to system capacity levels that are not less than
the load l in the system capacity pmf represented by the u-function U(z):
LOLP(l) = 0.0846 for 0<l≤3
LOLP(l) = 0.4947 for 3<l≤5
LOLP(l) = 0.5383 for 5<l≤6
LOLP(l) = 0.8258 for 6<l≤8
LOLP(l) = 1 for l>8
During the system’s operation time, the demand and reliability require-
ments can change. To provide a desired level of system performance,
management should develop a multistage expansion plan. For the problem
of optimal multistage system expansion, it is important to answer not only
the question of what must be included into the system, but also the question
of when it must be included into the system.
Example 3.6
Consider a power station coal transportation system consisting of five basic
components connected in series (Fig. 3.8):
1. subsystem of primary feeders;
2. subsystem of primary conveyors;
3. subsystem of stackers-reclaimers;
4. subsystem of secondary feeders;
5. subsystem of secondary conveyors.
4 5
1 3
2
The system belongs to the type of flow transmission system with flow
dispersion, since its main characteristic is the transmission capacity and
parallel elements can transmit the coal simultaneously. The system should
meet a variable demand W. Its acceptability function is defined as
F(G,W)=1( G ≥ W ). The system should have availability not less than
A*=0.99 for the given demand distribution w, q presented in Table 3.12.
Each system element is an element with total failure (which means that it
can have only two states: functioning with the nominal capacity or total
failure, corresponding to a capacity of zero). For each type of equipment, a
list of products available on the market exists. Each version of equipment is
characterized by its nominal capacity g, availability p, and cost c. The list
of available products is presented in Table 3.13. The maximal number of
elements in each component should not exceed six.
218 New Computational Methods in Power System Reliability
aj denotes the number of parallel elements of type i and version b: nib = aj.
One can see that a is a concatenation of substrings representing the vectors
n1, n2,…, nN.
Reliability of Generation Systems 221
Example 3.7
Consider the coal transportation system described in Example 3.6 and allow
each system component to consist of different versions of the elements.
The total number of available versions in the problem considered is
B = 7+5+4+9+4 = 29. Each string containing 29 integer numbers can
represent a possible solution. In order to illustrate the string decoding
process performed by the GA, consider the string
(0, 0, 0, 2, 0, 0, 1, 0, 0, 2, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 1, 0, 0).
This string corresponds to two primary feeders of version 4, one primary
feeder of version 7, two primary conveyors of version 3, one stacker of
version 1, three secondary feeders of version 7, and one secondary
conveyor of version 2.
According to step 2 of the decoding procedure, the u-functions of the
chosen elements are determined as u14(z) = 0.022z0+0.978z0.85, u17(z) =
0.016z0+0.984z0.26 (for the primary feeders), u23(z) = 0.003z0+0.997z0.53 (for
the primary conveyors), u31(z) = 0.029z0+0.971z1.00 (for the stacker), u47(z)
= 0.017z0+0.983z0.55 (for the secondary feeders) and u52(z) =
0.016z0+0.984z1.28 (for the secondary conveyor).
222 New Computational Methods in Power System Reliability
Example 3.8
Consider the same coal transportation system that was presented in
Example 3.6. The initial structure of this system is given in Table 3.16.
Each component contains a single subcomponent consisting of identical
elements: B'i = 1 for 1≤i≤5. The existing structure can satisfy the demand
presented in Table 3.12 with the availability A(w,q)=0.506. In order to
increase the system’s availability to the level of A*, the additional elements
224 New Computational Methods in Power System Reliability
The minimal cost system expansion solutions for different desired values
of system availability A* are presented in Table 3.17.
Consider, for example, the best solution obtained for A* = 0.99, encoded
by the string (0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0,
0, 0, 0, 0, 1). The corresponding minimal cost system expansion plan that
provides the system availability A = 0.990 presumes the addition of the
primary feeder of version 6, the primary conveyor of version 5, the stacker
of version 3, the secondary feeder of version 7 and the secondary conveyor
of version 4. The cost of this expansion plan is 4.358.
Reliability of Generation Systems 225
where IR is the interest rate. The total system expansion cost for the given
expansion plan {n1,…,nY} is
Y 1 N Bi
C= ∑ ∑ ∑ nib ( y)ci (b) (3.85)
y =1 (1 + IR )τ ( y ) i=1b=1
Having the system structure and demand distribution for each stage y,
one can evaluate the system performance measures. The problem of system
expansion optimization can be formulated as follows:
Find the minimal cost of system expansion plan that provides the
required level A* of the system availability Ay at each stage y:
C (n1,…,nY) → min
subject to f(Ay(n1,…,nY, w(y), q(y)), A*) =1 for 1≤ y≤Y (3.87)
where f(Ay,A*) is a function representing the desired relation between Ay
and A*.
The problem of system structure optimization can be considered as a
special case of this formulation. Indeed, if the set n' of the elements
composing the system at stage 0 is empty and Y=1, then the problem is
reduced to system structure optimization subject to reliability constraints. If
n' is empty but Y≠1, then the problem is that of multistage system structure
Reliability of Generation Systems 227
Example 3.9
In this example, we consider the same coal transportation system from
Example 3.6. The initial system structure is presented in Table 3.16. Table
3.18 contains the boiler system demand distributions at five different stages
and times from the present to the beginning of these future stages.
Because of demand increase, the availability index of the system
becomes A1(w(1), q(1)) = 0.506 at the first stage. To provide a desired
availability level at all the stages, the system should be expanded. The
characteristics of products available in the market for each type of
equipment are presented in Table 3.13.
Table 3.18. Demand distributions
Stage no. τ ( y) Demand distribution
1 0 w 1.00 0.80 0.50 0.20
q 0.48 0.09 0.14 0.29
2 3 w 1.20 0.80 0.50 -
q 0.43 0.32 0.25 -
3 6 w 1.40 1.20 0.80 0.50
q 0.10 0.39 0.31 0.20
4 9 w 1.40 1.20 0.80 -
q 0.41 0.24 0.35 -
5 12 w 1.60 1.40 1.00 -
q 0.30 0.45 0.25 -
One can compare expansion at the first stage of multistage plans with the
single-stage expansion plans presented in Table 3.17. The demand
230 New Computational Methods in Power System Reliability
distribution for the single-stage problem is the same as that in the first stage
of the multistage problem. The solutions for the first stage of the multistage
plan differ from ones that are obtained for single-stage expansion. The
comparison shows that a single-stage expansion plan is less expensive than
the first stage of a multistage plan, which should consider the effect of the
first stage expansion on the further stages.
To demonstrate the advantages offered by incremental expansion, the
solutions that satisfy reliability constraints for the final stage demand
distribution but require all expansion investments at time zero were
obtained (Table 3.22). One can see that incremental expansion provides
considerably less expensive solutions.
Each stage y (1≤y≤Y) begins τ(y) years after the base stage (stage 0) and is
characterized by its demand distribution w(y), q(y). At each stage, three
different types of action may be undertaken: modification of some of the
system’s elements, addition of new elements, and removal of some of the
system’s elements. In order to manipulate the system structure in a uniform
way, we will formulate all of these actions in terms of element
replacements:
The modification of element j of component i or changing the
procedure of the element maintenance leads to changes in its capacity
distribution. We can consider this action as replacement of the element of
version bij with the element of version b*ij. Parameters of new versions of
the element should be specified for each element modification action.
The addition of a new element into the system can be considered as the
replacement of the element of the corresponding type i and version 0 with a
new element with version b*ij. For each component i there are Bi element
versions available in the market.
The removal of the element of version bij from the system can be
formulated as its replacement with the element of version 0.
Now one can define each system modernization action x by vector
{ix, bix x , b*i x , cx} (3.91)
x
The total system modernization cost for the given modernization plan
x1, …, xY is
Y Y 1 H
C= ∑Cy = ∑ τ ( y) ∑ cx yh (3.93)
y=1 y=1 (1 + IR) h=1
Having the system structure Φ(y) and demand distribution w(y), q(y) for
each stage y, one can evaluate the system performance measures. The
problem of optimal system expansion can be formulated as follows (Levitin
and Lisnianski 1999):
Find the minimal cost system modernization plan that provides the
required level A* of the availability Ay at each stage y:
C (x1,…,xY) → min
subject to
f(Ay(x1,…, xy, w(y), q(y)), A*) =1 for 1≤ y≤Y. (3.94)
where f(Ay,A*) is a function representing the desired relation between Ay
and A*.
the elements composing the system at each stage y after the corresponding
replacements, one obtains the entire system capacity distribution by
applying the UGF technique. For this capacity distribution and for the
given demand distribution w(y), q(y) one obtains the system availability Ay.
The solution fitness for the estimated values of Ay and the total cost C
calculated using (3.93) is determined according to Equation (3.90).
Reliability of Generation Systems 233
Example 3.10
Consider two different optimization problems applied to the coal
transportation system from Example 3.6. In the first problem we have to
determine the initial system structure as well as its modernization plan. In
this case, the initial structure set is empty: Φ(0) = ∅. In the second problem
we have to find the optimal modernization plan for the system with the
initial structure presented in Table 3.23. Table 3.24 contains the demand
distributions at five different stages and times from the present to the
beginning of these future stages.
The demand distributions have the same maximal load at stages 1-3 and
4-5. Within these periods, variations of the load demand are caused by
intensification of boiler use. At stage 4, the installation of a new boiler is
planned, which should be supplied by the same coal supply system.
Because of increased demand, the availability index of the system
becomes A1(w(1), q(1)) = 0.506 at the first stage. To provide a desired
availability level A* = 0.95 at all of the stages, the system should be
modernized. The characteristics of the available modernization actions are
presented in Table 3.25 in the form (3.91). The action costs cx comprise
both the investment and operational costs.
costs for problems 1 and 2 are in the present values C = 1783.23 and C =
542.55 respectively.
Table 3.27. Optimal modernization plan for problem the with Φ(0) = ∅
Stage List of Structure of system components
No. actions A 1 2 3 4 5
1 1, 2*2, 2*6, 13, 0.958 6, 2*7 2*11 3, 18 3*20 3*23
14, 3*16, 3*19
2 15 0.951 6, 2*7 2*11 18, 19 3*20 3*23
Table 3.28. Optimal modernization plan for problem the with Φ(0) ≠ ∅
Stage List of Structure of system components
No actions A 1 2 3 4 5
1 6, 16, 0.950 2*1 3*2, 11 2*3 2*4, 20 2*26
2*21, 2*22
2 4, 2*15 0.951 1, 9 3*2, 11 2*19 2*4, 20 2*26
References
Allan R, Billinton R, Abdel-Gawad N (1986) IEEE reliability test system-
extensions to and evaluation of the generating system. IEEE Trans Power Syst,
1(4): 1-7.
Allan R, Billinton R (1992) Power system reliability and its assessment: Part 1.
Background and generating capacity. Power Engineering Journal, 6(4): 191-
196.
Ashworth M (1992) Integrated Demand and Supply Planning System with
Prosceen-II. In Williams, A. F. (Ed.) Energy Engineering, 89(3).
Babb R (1983) POWRSYM Production Costing Program, TVA.
Baleriaux H, Jamoulle E, Guertechin L (1967) Simulation de l’exploitation d’un
de machines thermiques de production d’electricite couple a des station de
pompage. Revue (edition SRBE) 5: 225-245.
Billinton R and Li W (1992) Consideration of multi-state generating unit models in
composite system adequacy assessment using Monte Carlo simulation.
Canadian Journal of Electrical and Computer Engineering 17(1).
Billinton R and Li W (1994) Reliability Assessment of Electric Power System
Using Monte Carlo Methods. Plenum Press, New York.
Billinton R and Wang P (1999) Deregulated power system planning using a
reliability network equivalent technique. IEE Proc. Gener. Transm. Distrib.
146(1): 25-30.
Booth R (1972) Power system simulation model based on probability analyses.
IEEE Transactions on Power Apparatus and System 9: 62-69.
Ding Y, Wang P and Lisnianski A (2006) Optimal reserve management for
restructured power generating systems. Reliability Engineering and System
Safety 91: 792-799.
Fleck W, Charny L (1984) Electric generation expansion analysis system. Final
report. EPRI-EL-2561-6, Stone and Webster Engineering Corp., Boston, MA
(USA)
Gnedenko B, Ushakov I (1995) Probabilistic reliability engineering. John Wiley
and Sons. New York.
Gurevich V, Or O, Frant Sh. (1992) Maintenance scheduling optimization
Program. Internal Report RD-572, Israel Electric Corp.
Gurevich V, Frant Sh, Bianu A (1995). A generation system reliability model with
Monte Carlo simulation. Proceedings of 18th IEEE convention in Israel.
IEEE Task Force. (1979) IEEE reliability test system. IEEE Trans. Power.
Apparatus Syst. PAS-98: 2047-54.
Jenkins R, Joy D (1974) WIEN Automatic system planning package (WASP).
Oak-Ridge National Laboratory.
Levitin G, Lisnianski A, Elmakis D (1997) Structure optimization of power system
with different redundant elements. Electr Pow Syst Res 43: 19-27.
Levitin G, Lisnianski A, Ben-Haim H, Elmakis D (1998) Redundancy optimization
for series-parallel multi-state systems. IEEE Trans Reliab 47: 165-72.
Levitin G, Lisnianski A (1999) Optimal multistage modernization of power system
subject to reliability and capacity requirements. Electr Pow Syst Res 50: 183-
90.
238 New Computational Methods in Power System Reliability